پیش بینی ساختاری برای اطلاعات بازاریابی   Structural forecasts for marketing data

1. Introduction The functions of marketing within an organization are to represent existing and potential consumers and to help to guide product policy, including the development of new and existing goods and services, pricing, communication, and distribution activities within the firm. Marketing’s interest in forecasting is related primarily to the prediction of the effects of various actions in terms of sales and demand. Examples of this include the effects of new product features on demand, changes in the pricing structure of an offering, advertising initiatives that aim to build brand recognition, and changes to channel policy, ranging from the use of existing dealerships to new electronic and digital venues. Because of the complexity of human behavior, marketing is also interested in predicting the effects of its expenditures on other variables that eventually lead to sales, such as brand recognition, recall, satisfaction and purchase intent. Marketing forecasts are often context-dependent and disaggregate in nature. Forecasts pertain to specific brands in specific geographic regions, are designed to considerspecific aspects of seasonality (e.g., fourth of July) and specific consumption occasions (e.g., backyard picnics), and are targeted at specific types of individuals. This is not to say that aggregate predictions of sales are unimportant to marketing, but that one of the primary goals of marketing is to forecast a demand that may not yet exist, which is generated from individuals who may not yet participate in the product category. It is important to understand the source of increased sales, because firms may have multiple offerings within a product category and wish to gain shares from specific competitors instead of from their own brands. Some marketing interventions are oriented toward growing the market, while others are merely reactions to the competition. The result of these factors is that assuming a naïve model for demand (e.g., an aggregate exponential smoothing model) is often inadequate. The micro-foundation of the forecast is important because the interventions being considered are thought to have specific effects on the consumer utility and sales. Marketing forecasts are challenged by the nature of marketing data and the people from whom the demand is generated. People are heterogeneous in their preferences and sensitive to marketing variables such as prices. Their purchases are represented by ‘lumpy’ data, where the most frequently observed number is zero. The data cubein marketing that corresponds to (respondents × products × time) is sparsely populated, in the sense that most people do not buy most products, visit most websites or attend to most attempts to get their attention. In fact, it is so costly to change an individual’s nature that an overarching maxim in marketing is to ‘‘make what people will want to buy’’, not just to attempt to ‘‘make them want to buy’’. The issue of limited data at the individual level means that both point estimates and measures of the uncertainty are needed for forecasting, and the aim of understanding the determinants of the demand as well as their implications for market share and profitability means that traditional loss functions, such as the mean squared error of sales, are not always appropriate.